Corpus

Introduction

Coffee - Road

This project concerns Spotify’s ‘Coffeehouse’ and ‘On the Road’ playlists. A coffeehouse is a pleasant place where you can talk, work or read while enjoying a cup of coffee; you will probably listen to easy, mellow songs here. But, when you’re on the road, you have to kill time while going from A to B - whether you’re alone or with friends; you go to work or are on a holiday. Since the car is a private place, you might listen to familiar, sing-along songs. These settings give a completely different vibe, so my goal is to see whether Coffeehouse and On the Road playlists have distinctive features, and in what ways they differ.

Popular Coffeehouse and On the Road playlists were selected, resulting in 271 Coffeehouse and 250 On the Road songs. The distribution of keys and mode can be seen in the plots - the lower part represents minor songs; the top major.

Playlists {data-width = 700}

Playlists

The following playlists were used for this research, composed by Spotify, Filtr Nederland and Digster Nederland:

Coffeehouse:

1. Coffeehouse - Feel the vibe of the Coffeehouse playlist, full of relaxing singer-songwriter an pop music.
2. 't Koffiehuis - 'n Bakkie en gemoedelijke muziek op de achtergrond.    
3. Your Favorite Coffeehouse - Curl up in your favorite spot with some sweet, mellow tunes...

On the Road:

1. Classic Road Trip Songs - The ultimate playlist to fuel your good mood while on the road.
2. Onderweg - Met deze playlist heb je altijd de beste hits bij de hand voor onderweg.
3. Road Trip - Go on a weekend getaway with your barkada and your favorite hits!
4. Songs to Sing in the Car - Sing along and enjoy the drive...

Keys Coffeehouse {data-height = 250}

Keys Coffeehouse

Keys On the Road {height = 200}

Keys On the Road

Findings

Text

Track popularity

The Coffeehouse playlists have a mean track popularity of 52.38 (SD = 25.75), whereas the On the Road playlists have a higher mean track popularity of 64.07 (SD = 22.07). This makes sense, since the On the Road playlists contain more songs that everyone can sing along to, and are thus familiar to a large audience.

Energy

As expected, energy in On the Road playlists is much higher than for Coffeehouse playlists. While on the road, you need to stay awake, so most songs have a loud beat. Coffeehouse songs, on the other hand, sound more quiet and peaceful. Spotify measured that the average energy in On the Road songs is 0.6123 (SD = 0.18) and 0.3596 (SD = 0.26) in Coffeehouse songs. Indeed, energy seems to be much higher in On the Road playlists.

Acousticness

Characteristic for mellow Coffeehouse songs is their high acousticness (0.6589, SD = 0.35). The playlists contain mostly easy, soft songs that are accompanied by piano or guitar without much of electronic sounds, and even contain a few acoustic versions The acousticness of these playlists is thus much higher than for On the Road playlists (0.2813, SD = 0.26).

Valence

I experience the On the Road playlists to be happier than Coffeehouse playlists, since these songs sound more uplifting to me. Spotify has generated a measure for valence: the higher the valence value, the happier the song. According to this measure, On the Road playlists are indeed slightly happier (0.4755, SD = 0.21) than Coffeehouse playlists (0.3555, SD = 0.32).

Boxplots

All-in-one

### Plot

Sidebar

All-in-one

So, high acousticness seems to be characteristic for Coffeehouse songs, whereas high energy is characteristic for Road Trip songs. Valence is higher in Coffeehouse playlists, and track popularity is higher in On the Road playlists. These four values are visualised in one plot to see a clear difference. Most of the Coffeehouse songs are in the left upper corner and On the Road songs are in the right lower corner. This figure indeed shows mostly high acoustiness and low energy in Coffeehouse, and high energy and low acousticness in On the Road. Songs with high energy and low acoustiness are happiest, but track popularity is not really clear.

(By the way, ggplotly makes it really hard to add legends: size = valence, transparency = track popularity)

Random Forest classification

Figures

Features

Classification

Sidebar

Random Forest

The Random Forest Classification predicts whether random songs are from a Coffeehouse or On the Road playlist, with an accuracy of approximately 80%. We already saw that high acousticness and high energy are characteristic for Coffeehouse and On the Road songs, respectively, and next to that, the timbre features c01 (loudness), c02 (brightness) and c04 (strong attack), seem to be the most meaningful when making the prediction. That’s why I chose to let the model classify with just these five features. Since the prediction is based on only five features, accuracy drops to approximately 70%. That’s not really good, but these features approach a fair prediction between Coffeehouse and On the Road songs.

Two songs

Text

Analysis

A good example of a Coffeehouse song would be Waiting Around (Aisha Badru). It is a pleasant, mellow song with high acousticness (0.988). On the other hand, Walk of Life is a typical On the Road song: it is the song with most energy (0.960) and I think everyone can sing along. I created a chromagram and a Self-Similarity Matrix based on timbre for both songs.

According to Spotify’s API, Waiting Around is in F major: F-A-C is indeed found in the chromagram, although it might as well be in C major. Walk of Life shows mostly E and B, but a wide range of pitch classes can be found. E and B conform with Spotify saying it is in E major.

In Waiting Around, I can identify the chorus around 50 and 100 seconds, and the bridge must be around 130 seconds. Walk of Life shows a completely different pattern. It seems to repeat itself throughout the whole song, which I can hear as well. The chorus is around 100, 150 and 200 seconds. There is no bridge, but an instrumental fade-out at the end.

However, I don’t think that an anlysis of just two songs is representative for the whole corpus. Not all songs should have the same pattern as is shown here.

Plots

Chromagram Waiting Around

Chromagram Walk of Life

Self-Similarity Matrix (timbre) Waiting Around

Self-Similarity Matrix (timbre) Walk of Life

Individual Clustering

Row

Coffeehouse

On the Road

Sidebar

Individual Clustering

These visuals show what songs Spotify considers to belong to the same group within the playlists. These figures only cover 50 songs of both playlists, because a dendrogram of the whole corpus is visually too complex.

The first thing I notice, is that Coffeehouse a cluster that consists of only one song, Walls (Kings of Leon). I’m not sure why, because it is a good Coffeehouse song. Coffeehouse also seems to have smaller sub-groups: the right-hand side keeps getting smaller branches. On the Road clearly has two bigger subgroups that consist of half the playlist.

Combined Clustering

Text

Analysis

When combining a few songs of both playlists, the model is not doing a bad job clustering, but it’s not perfect either. The pink color represents On the Road songs, blue Coffeehouse. We can see then, that it groups Coffeehouse songs together, including a few lost On the Road songs. There are not individual On the Road clusters, which is weird. Apparently, the chosen songs have something in common with the Coffeehouse songs.

And again, Walls has it’s own branch. Looking at the heat map, its instrumentalness is way higher than all the other songs, maybe that’s why it stands out.

When looking at the heat map, there are no really outstanding features that catch my eye (no big yellow parts). There is a small cluster around speechiness, and I can understand why songs with a lower volume and low energy are grouped together. Also, I think a group is formed around high energy and low acousticness. Liveness has a cluster as well, but it’s odd because those songs arent’ actually live recordings.

Visuals

Dendrogram

Heatmap

[1] 432.0 345.6

“Outliers” & conclusion

Text

So…?

I noticed that a few songs appeared in both a Coffeehouse and On the Road playlist. If some songs can be both from a Coffeehouse or On the Road playlist, does it mean that it’s still hard to distinguish a Coffeehouse song from a On the Road song? I can understand why these songs were added to both playlists, but please hear for yourselves! Would you listen to these songs in a coffeehouse or on the road?

Apart from that, I think acousticness and energy appear to be the most distinctive characteristics of Coffeehouse and On the Road playlists. A model that is trained to classify random songs as Coffeehouse or On the Road song cannot be totally trusted, and clustering is also not really effective. It’s not the case that Coffeehouse songs cannot be listened in the car or vice versa: if the user wants to hear more acoustic and mellow songs, a Coffeehouse playlist is useful, whereas an On the Road playlists contains more energetic and sing-along songs. Besiders, there is of course still the option to combine the best of two worlds. I was just wondering whether certain charactersitics are available for creating uch different atmospheres, and apparently acousticness and energy are to a certain extent in this case.

And, of course, I learned some basics of doing research in Computational Musicology, and gained some insight about all the possibilities with Spotify’s API. I have enjoyed following this course and working on the project, and hope to work with Spotify’s features in further research project!